Engineering machinery knowledge graph data enhancement method and system

Abstract

The embodiment of the invention provides an engineering machinery knowledge graph data enhancement method and system, and belongs to the technical field of knowledge graphs. The method comprises the following steps: performing engineering machinery related entity identification based on text information, performing similar entity generation based on each identification entity, and constructing a similar entity set corresponding to each identification entity; calculating the average similarity of the corresponding identification entities based on the similar entity set corresponding to each identification entity; calculating the probability of occurrence of each identification entity in the text information based on a worming technology, and reserving the identification entities with the entity evaluation values greater than a preset evaluation value threshold and the corresponding similar entities; and executing engineering machinery knowledge graph updating based on the reserved identification entities and the corresponding similar entities. According to the scheme, the problem of sparse graph content when data is insufficient in the traditional technology is effectively solved, the integrity and depth of the knowledge graph are improved, and the application value of the knowledge graph in the field of engineering machinery is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of knowledge graph technology, and specifically to a method for enhancing engineering machinery knowledge graph data and a system for enhancing engineering machinery knowledge graph data. Background Art

[0002] As a structured semantic knowledge base, the knowledge graph enables users to obtain more accurate query results through the relationships between nodes and the attributes of the nodes themselves. When constructing a knowledge graph, the existing technology mainly relies on entity extraction and relationship extraction of input data, adding the identified entities and relationships to the graph to complete the construction of the knowledge graph. However, when applied to the field of engineering machinery, the existing solutions expose several technical problems.

[0003] The knowledge in the field of engineering machinery is highly specialized and complex. Existing technologies rely heavily on the quality and quantity of input text when identifying entities and extracting relationships. However, the knowledge in the field of engineering machinery is often difficult to collect comprehensively, resulting in limited entities and relationships available for extraction. This situation directly leads to the content in the knowledge graph being too sparse, the number of nodes being insufficient, and the number of node attributes being too few, making it impossible to effectively represent the rich knowledge in the field of engineering machinery. Due to the lack of graph content, the existing knowledge graph cannot meet the needs of practical applications in terms of depth and breadth. This problem of insufficient depth makes it difficult for knowledge graphs to provide accurate search results when processing complex queries, thereby limiting their application value in the field of engineering machinery. In addition, the small amount of data also makes it difficult to build a knowledge graph of sufficient scale, which further weakens the applicability of the graph in practical scenarios.

[0004] Therefore, in the process of constructing knowledge graphs in the field of engineering machinery, the existing technology has problems such as small graph scale, incomplete node attributes, and insufficient depth of graph content. There is an urgent need for a technical solution that can effectively improve the integrity and depth of graph content to meet the high requirements of the knowledge graph in the field of engineering machinery. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a method and system for data enhancement of a knowledge graph of engineering machinery, so as to at least solve the problems of small graph scale, incomplete node attributes and insufficient graph content depth in the prior art in the process of constructing knowledge graphs in the field of engineering machinery.

[0006] In order to achieve the above-mentioned purpose, the first aspect of the present invention provides a method for data enhancement of a knowledge graph of construction machinery, the method comprising: collecting text information of construction machinery to be stored in the database, identifying construction machinery-related entities based on the text information and generating similar entities based on each identified entity, and constructing a similar entity set corresponding to each identified entity; calculating the average similarity of the corresponding identified entity based on the similar entity set corresponding to each identified entity; calculating the occurrence probability of each identified entity in the text information based on the word embedding technology, and obtaining an entity evaluation value based on the occurrence probability and the average similarity of the corresponding identified entity, retaining the identified entities and their corresponding similar entities whose entity evaluation values ​​are greater than a preset evaluation value threshold; and updating the knowledge graph of construction machinery based on the retained identified entities and their corresponding similar entities.

[0007] Optionally, the identifying engineering machinery related entities based on the text information includes: performing text cleaning on the text information, and performing word segmentation processing after text cleaning to obtain multiple word segmentation items; wherein, based on a pre-built engineering machinery knowledge graph, a corresponding engineering machinery entity recognition model is loaded, and based on the loaded engineering machinery entity recognition model, engineering machinery related entity recognition is performed in each word segmentation item to obtain a recognized entity.

[0008] Optionally, the rules for generating similar entities based on each identified entity are as follows: S1) taking each identified entity as an input item, loading a pre-trained engineering machinery generation model, obtaining multiple generated entities, and assigning a corresponding similarity probability to each generated entity; S2) comparing the similarity probabilities of each generated entity, and taking the previous generated entity with the largest similarity probability as the similar entity for this round of training; S3) deleting the determined similar entity from the entities to be generated, repeating steps S1)-S2) N times, and obtaining N similar entities.

[0009] Optionally, based on the similarity between each similar entity in the similar entity set and the corresponding identified entity, similar entities whose similarity is less than a preset similarity threshold are filtered out, including: converting each similar entity in the similar entity set and the identified entity into vectors of preset same dimension; calculating the similarity between the original entity and each generated similar entity based on cosine similarity; comparing the size relationship between the corresponding similarity of each similar entity and a preset similarity threshold, and filtering similar entities whose similarity threshold is less than the preset similarity threshold.

[0010] Optionally, the occurrence probability of each recognized entity in the text information is calculated based on word embedding technology, including: performing word segmentation processing on the text information to determine the position of each recognized entity in the text information to obtain context word entities; loading pre-trained word embedding technology to process each context word entity and the corresponding recognized entity into a vector of a preset dimension; calculating the association between each recognized entity and each context word based on each context word entity vector and the corresponding recognized entity vector; counting the occurrence frequency of each recognized entity in the text type of the text information based on a pre-constructed engineering machinery knowledge graph; calculating the occurrence probability of each recognized entity in the text information based on the association and the occurrence frequency; when calculating the occurrence probability of each recognized entity in the text information based on word embedding technology, the method also includes: filtering out recognized entities whose occurrence probability is less than a preset probability threshold, and similar entities corresponding to the recognized entities.

[0011] Optionally, a weighted summation operation is performed based on the occurrence probability and the average similarity of the corresponding identified entities, including: weighting the occurrence probability and the average similarity of each identified entity based on a preset weighting scheme; traversing all identified entities, and performing a weighted summation operation based on the weighted occurrence probability and the weighted average similarity, respectively, to obtain an entity evaluation value corresponding to each identified entity.

[0012] Optionally, the updating of the engineering machinery knowledge graph based on the retained identified entities and their corresponding similar entities includes: when it is determined that there is no corresponding identified entity in the pre-built engineering machinery knowledge graph, calculating the correlation between each identified entity and the corresponding similar entity and the corresponding entity type in the pre-built engineering machinery knowledge graph; classifying each identified entity and the corresponding similar entity based on the correlation, and constructing graph nodes for each identified entity and the corresponding similar entity based on the classification results; after determining that the graph node meets the node standard, updating the engineering machinery knowledge graph based on the constructed graph node.

[0013] Optionally, classification is performed on each identified entity and the corresponding similar entity based on the correlation, including: determining the entity type of the corresponding identified entity / similar entity based on the entity type corresponding to the maximum value of the correlation between each identified entity / similar entity and each entity type in the pre-built engineering machinery knowledge graph; wherein the entity type is a node entity or a node attribute.

[0014] Optionally, the method performs graph node construction on each identified entity and the corresponding similar entities based on the classification results, including: calculating the correlation between each node entity and the corresponding node attributes, and taking the combination with a correlation greater than a preset correlation threshold as a node combination; extracting a node construction template, performing graph node construction based on each node combination, and obtaining multiple candidate graph nodes.

[0015] Optionally, after determining whether the graph node meets the node standard, the construction machinery knowledge graph is updated based on the constructed graph node, including: determining whether each candidate graph node has missing information based on the node construction template; if there is missing information, generating the missing information based on the existing information and the pre-trained node generation model, and updating each candidate graph node based on the generated missing information; treating the candidate graph nodes that meet the node construction template as standard graph nodes, and updating the construction machinery knowledge graph based on the standard graph nodes.

[0016] The second aspect of the present invention provides a construction machinery knowledge graph data enhancement system, the system comprising: a collection unit, used to collect text information of construction machinery to be stored in the database, identify construction machinery-related entities based on the text information and generate similar entities based on each identified entity, and construct a similar entity set corresponding to each identified entity; a similarity determination unit, used to calculate the average similarity of the corresponding identified entity based on the similar entity set corresponding to each identified entity; an evaluation unit, used to calculate the occurrence probability of each identified entity in the text information based on the word embedding technology, and obtain an entity evaluation value based on the occurrence probability and the average similarity of the corresponding identified entity, retaining the identified entities and their corresponding similar entities whose entity evaluation values ​​are greater than a preset evaluation value threshold; an update unit, used to perform construction machinery knowledge graph update based on the retained identified entities and their corresponding similar entities.

[0017] On the other hand, the present invention provides a computer-readable storage medium, which stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned engineering machinery knowledge graph data enhancement method.

[0018] Through the above technical scheme, the scheme of the present invention collects the text information of engineering machinery to be stored, performs relevant entity recognition and similar entity generation, constructs a similar entity set, and screens and weighted sums based on the probability of occurrence calculated by similarity and word embedding technology to obtain the entity evaluation value. The identified entities and similar entities with evaluation values ​​greater than the preset threshold are retained and ultimately used to update the engineering machinery knowledge graph. The scheme of the present invention effectively overcomes the problem of sparse graph content in traditional technologies when data is insufficient. By introducing the generation and screening of similar entities, the integrity and depth of the knowledge graph are improved, and its application value in the field of engineering machinery is enhanced.

[0019] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0021] Figure 1 It is a flowchart of the steps of a method for enhancing data of a knowledge graph of engineering machinery provided by one embodiment of the present invention;

[0022] Figure 2 It is a flow chart for implementing a method for enhancing engineering machinery knowledge graph data provided by one embodiment of the present invention;

[0023] Figure 3 It is a system structure diagram of an engineering machinery knowledge graph data enhancement system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0025] Figure 1 1 is a method flow chart of a method for enhancing engineering machinery knowledge graph data provided by an embodiment of the present invention. Figure 1 As shown, an embodiment of the present invention provides a method for enhancing data of a knowledge graph of engineering machinery, the method comprising:

[0026] Step S10: Collect text information of engineering machinery to be stored, identify engineering machinery related entities based on the text information, and generate similar entities based on each identified entity to obtain multiple similar entities of each identified entity, and construct a similar entity set corresponding to each identified entity.

[0027] Specifically, the identification of engineering machinery related entities based on the text information includes: performing text cleaning on the text information, and performing word segmentation processing after text cleaning to obtain multiple word segmentation items; wherein, based on the pre-constructed engineering machinery knowledge graph, the corresponding engineering machinery entity recognition model is loaded, and based on the loaded engineering machinery entity recognition model, engineering machinery related entity recognition is performed in each word segmentation item to obtain a recognized entity.

[0028] In an embodiment of the present invention, in text processing in the field of engineering machinery, building and maintaining high-quality knowledge graphs is essential to support intelligent query and information extraction. In order to achieve this goal, it is first necessary to effectively process and identify the engineering machinery text information. This process includes several key steps: text cleaning, word segmentation, entity recognition, etc. The purpose of text cleaning is to remove irrelevant or noisy information in the text, such as extra spaces, useless punctuation, and non-standard characters that may affect subsequent analysis. By cleaning the text, it can be ensured that the subsequent processing steps can accurately identify and analyze the text content. This step not only improves the standardization of the text, but also enhances the accuracy and efficiency of the model when processing text.

[0029] Furthermore, after the text is cleaned, the next step is word segmentation. Word segmentation is to divide the text into independent words or phrases (i.e., word segmentation items), which constitute the basic analysis unit of the text. Through effective word segmentation, the potential information in the text can be expressed in a more structured way, laying the foundation for subsequent entity recognition.

[0030] Furthermore, after the word segmentation is completed, the corresponding engineering machinery entity recognition model is loaded based on the pre-built engineering machinery knowledge graph. As a structured semantic knowledge base, the knowledge graph contains a large number of entities and their relationships in the field of engineering machinery. These entities and relationships can be effectively utilized through the entity recognition model in practical applications. The process of loading the entity recognition model depends on the information in the knowledge graph, so that the model can recognize the entities related to engineering machinery in the text. Based on the loaded engineering machinery entity recognition model, the engineering machinery related entities are identified in each segmentation item obtained by the word segmentation process. Entity recognition is to determine whether the segmentation item corresponds to an engineering machinery entity by analyzing the matching degree of each segmentation item with the known entities in the knowledge graph. For example, after the input text "the boom of the pump truck cannot move during the pumping process" is recognized through entity recognition, the key entities in it can be identified as "pump truck", "boom", "pumping" and "cannot move". These entities represent important concepts and operating states in the field of engineering machinery, and are the information that should be stored and managed in the knowledge graph.

[0031] Based on the solution of the present invention, text cleaning and word segmentation processing can effectively remove noise and useless information, so that the entity recognition model can more accurately match and identify entities related to construction machinery, reducing the probability of misrecognition. Loading the entity recognition model based on the pre-built knowledge graph enables the model to use rich domain knowledge for recognition, enhancing the accuracy and relevance of recognition. This recognition method based on domain knowledge is more effective than the traditional pure data-driven method. Through efficient entity recognition, a large number of engineering machinery-related entities contained in the text can be extracted and added to the knowledge graph. This not only enriches the content of the graph, but also provides more comprehensive and in-depth information support for subsequent intelligent query and analysis.

[0032] Furthermore, the rules for generating similar entities based on each identified entity are as follows: S1) taking each identified entity as an input item, loading a pre-trained engineering machinery generation model, obtaining multiple generated entities, and assigning a corresponding similarity probability to each generated entity; S2) comparing the similarity probabilities of each generated entity, and taking the previous generated entity with the largest similarity probability as the similar entity for this round of training; S3) deleting the determined similar entity from the entities to be generated, and repeating steps S1)-S2) N times to obtain N similar entities.

[0033] In an embodiment of the present invention, first, the identified engineering machinery entity is used as an input item and input into a pre-trained engineering machinery generation model. This generation model is obtained after fine-tuning GPT-3. After being trained specifically for data in the field of engineering machinery, the generation model can generate multiple similar entities related to the input entity. Each generated entity is assigned a corresponding similarity probability, which represents the semantic proximity between the generated entity and the input entity.

[0034] Furthermore, after generating multiple similar entities, the similarity probabilities of these entities are compared, and the entity with the highest similarity probability is taken as the similar entity generated in this round. In this way, it is ensured that the generated similar entities have the highest relevance to the original recognized entities. This step effectively filters out those generated entities that have a low relevance to the original entities, ensuring that the new nodes added to the knowledge graph have high accuracy and relevance.

[0035] Furthermore, in order to obtain more high-quality similar entities, the identified similar entities are deleted from the entities to be generated to avoid repeated generation of the same entity. Then, steps S1 and S2 are repeated N times to obtain N different similar entities. For example, if the generation model is set to generate 5 results each time and repeated 10 times, 10 similar entities can be obtained for each identified entity. This method not only ensures the diversity of similar entities, but also optimizes the quality of generated entities through multiple screenings.

[0036] Based on the scheme of the present invention, through multiple rounds of screening of entities with the highest similarity probability, the method ensures that the generated similar entities are highly semantically associated with the original identified entities. This high-precision generation of similar entities effectively avoids the addition of irrelevant entities and improves the overall quality of the knowledge graph. Multiple generation and screening processes enable each identified entity to obtain multiple high-quality similar entities. These similar entities not only expand the number of nodes in the knowledge graph, but also enhance the complexity and hierarchy of the graph, providing richer semantic associations for subsequent intelligent analysis and applications.

[0037] Furthermore, since the construction machinery generation model has been fine-tuned with domain knowledge, the similar entities it generates have stronger adaptability and accuracy in the field of construction machinery. Compared with general models, this domain-specific model can more accurately capture the special semantics of the construction machinery field and improve the professionalism of the generated results. By generating highly correlated similar entities, the knowledge graph can better support intelligent applications, such as recommendation systems, fault diagnosis, and maintenance recommendations for construction machinery equipment. These applications rely on the accuracy and comprehensiveness of the knowledge graph, and the generation of similar entities is one of the key factors to ensure the success of these intelligent applications.

[0038] Step S20: calculating the average similarity of the corresponding identified entities based on the similar entity sets corresponding to each identified entity.

[0039] Specifically, each similar entity and the identified entity in the similar entity set is converted into a vector of preset same dimension; the similarity between the original entity and each generated similar entity is calculated based on cosine similarity; the size relationship between the similarity corresponding to each similar entity and the preset similarity threshold is compared, and the similar entities whose similarity threshold is less than the preset similarity threshold are filtered.

[0040] In the embodiment of the present invention, Figure 2In order to ensure a high degree of correlation between entities and similar entities in the graph, the generated similar entities must be further screened and optimized. The specific steps include converting each similar entity and the original identified entity into vectors of the same preset dimension, and using cosine similarity to calculate similarity based on these vectors. Cosine similarity, as a commonly used similarity measurement method, can effectively measure the angular relationship between two vectors. The value range is between 0 and 1. The closer the value is to 1, the higher the similarity between the two vectors. After calculating the similarity between the original entity and each generated similar entity, the similarity of each similar entity is compared with the preset similarity threshold. Only those similar entities with a similarity higher than the threshold will be retained and enter the next step of the knowledge graph. For example, if the similarity threshold is set to 0.8, the similarity between "arm" and "one arm" is 0.9, indicating that they are highly semantically related and therefore retained; while the similarity between "arm" and "leg" is only 0.01, and the correlation is extremely low, so it is filtered out.

[0041] Based on the scheme of the present invention, the semantic correlation between entities and similar entities in the engineering machinery knowledge graph is significantly improved through vectorization processing and cosine similarity calculation. Through a strict similarity screening mechanism, it is ensured that the entity nodes and similar entity nodes in the graph have a high degree of semantic association, thereby enhancing the accuracy and practicality of the knowledge graph. Entities with high similarity are retained, making the structure of the graph more reasonable and the content richer, thereby improving the application effect of the knowledge graph in intelligent retrieval, fault diagnosis and domain knowledge reasoning. Ultimately, the screening and filtering process lays a solid foundation for building a high-quality and reliable knowledge graph, ensuring the data accuracy and decision-making effectiveness in application scenarios in the field of engineering machinery.

[0042] Furthermore, the average similarity of the corresponding identified entities is calculated based on the remaining similar entities. After some similar entities with high similarity to the original identified entities are retained through similarity screening, the next step is to calculate the average similarity of the corresponding identified entities based on these remaining similar entities. This average similarity is obtained by summing up the similarity values ​​of all retained similar entities and then dividing it by the number of similar entities. The process of calculating the average similarity not only helps to further quantify the overall association between the identified entity and its similar entities, but also provides a comprehensive similarity measure for subsequent knowledge graph updates and optimizations. For example, if the average similarity of an identified entity is high, it means that its set of similar entities is highly semantically consistent with the original entity. Such entities will be given priority for addition to the knowledge graph to enhance the semantic hierarchy and accuracy of the graph.

[0043] Step S30: Calculate the occurrence probability of each recognized entity in the text information based on the word embedding technology, and obtain the entity evaluation value based on the occurrence probability and the average similarity of the corresponding recognized entity, and retain the recognized entities whose entity evaluation values ​​are greater than the preset evaluation value threshold and their corresponding similar entities.

[0044] Specifically, the occurrence probability of each recognized entity in the text information is calculated based on the word embedding technology, including: performing word segmentation processing on the text information, determining the position of each recognized entity in the text information, so as to obtain context word entities; loading pre-trained word embedding technology, processing each context word entity and the corresponding recognized entity into a vector of a preset dimension; calculating the association between each recognized entity and each context word based on each context word entity vector and each corresponding recognized entity vector; counting the occurrence frequency of each recognized entity in the text type of the text information based on a pre-constructed engineering machinery knowledge graph; calculating the occurrence probability of each recognized entity in the text information based on the association and the occurrence frequency; in calculating the occurrence probability of each recognized entity in the text information based on the word embedding technology, the method also includes: filtering out recognized entities whose occurrence probability is less than a preset probability threshold, and similar entities corresponding to the recognized entities.

[0045] In an embodiment of the present invention, a word segmentation process is performed on the text information to determine the specific location of each recognized entity in the text. This step can effectively identify all possible entities in the text and extract the context words of these entities. The context word entities are the words around the recognized entities, which together constitute the semantic environment of the recognized entities. The relationship between these context words and the recognized entities is crucial for calculating the occurrence probability of the recognized entities.

[0046] In a possible implementation, the word embedding technology selected is to use the word2vec model to calculate the probability of occurrence of each recognized entity in the text information. The pre-trained word2vec model is loaded, and each context word entity and the corresponding recognized entity are converted into a vector of a preset dimension. The word2vec model can capture the semantic relationship between words and map these relationships into a vector space. This vectorized representation provides a basis for subsequent correlation calculations. By processing context words and recognized entities as vectors, the semantic correlation between them can be accurately calculated using mathematical means.

[0047] Furthermore, after vectorization, the correlation between each recognized entity and the context word is calculated based on the entity vector of each context word and the corresponding recognized entity vector. The correlation reflects the semantic adaptability of the recognized entity in its context and can help determine the importance and relevance of the recognized entity in the text. A high correlation indicates that there is a strong semantic association between the recognized entity and its context words, while a low correlation indicates that the recognized entity may not be very relevant to the context. Based on the pre-built engineering machinery knowledge graph, the frequency of occurrence of each recognized entity in the text information is counted. The knowledge graph contains a lot of domain knowledge, and frequency statistics help to further quantify the importance of recognized entities. Combining the correlation and the frequency of occurrence, the probability of occurrence of each recognized entity in the text information can be more accurately calculated. This probability takes into account the semantic environment and frequency of occurrence of the entity in the text, thereby providing a more accurate measure of entity importance.

[0048] Furthermore, after calculating the probability of occurrence, the system will filter out those recognized entities and their similar entities whose probability of occurrence is less than a preset threshold. Through this filtering, entities with low relevance or low frequency of occurrence in the text can be removed, ensuring that the retained entities are highly semantically relevant to the text content.

[0049] Based on the scheme of the present invention, through word segmentation, vectorization, correlation calculation and frequency statistics, the importance of each recognized entity in the text can be accurately measured, and the quality of entities in the knowledge graph can be ensured through probability threshold screening. Ultimately, this process not only enhances the semantic level of the knowledge graph, but also improves its practical utility in intelligent applications in the field of engineering machinery. By filtering irrelevant or less important entities, the knowledge graph can better support applications such as complex queries, intelligent recommendations and semantic reasoning, significantly improving the efficiency and accuracy of information processing in the field of engineering machinery.

[0050] Furthermore, a weighted summation operation is performed based on the occurrence probability and the average similarity of the corresponding identified entities, including: weighting the occurrence probability and the average similarity of each identified entity based on a preset weighting scheme; traversing all identified entities, and performing a weighted summation operation based on the weighted occurrence probability and the weighted average similarity, respectively, to obtain an entity evaluation value corresponding to each identified entity.

[0051] In an embodiment of the present invention, based on a preset weighting scheme, weights are assigned to the occurrence probability and average similarity of each identified entity. The purpose of weighting is to adjust the weights of the occurrence probability and similarity in the final evaluation according to different application scenarios and field requirements. For example, in some applications, the similarity of entities may be more important, so the weight of similarity will be higher, while in other cases, the occurrence probability of entities may account for a larger proportion. Through this flexible weighting mechanism, the needs of different scenarios can be better met.

[0052] Furthermore, all identified entities are traversed, and based on the weighted occurrence probability and the weighted average similarity, weighted sum operations are performed to calculate the entity evaluation value of each identified entity. For example: weighted result = 0.2*occurrence probability + 0.8*average similarity. The weights (0.2 and 0.8) in this calculation formula can be adjusted according to actual needs to achieve the best screening effect.

[0053] Preferably, the entity evaluation value is a comprehensive measure of the quality of the recognized entity, reflecting the importance of the entity in the text and its semantic association with similar entities. When the weighted result is greater than a preset threshold (such as 0.6), it means that the entity is highly semantically related to the text content and has a high retention value, so the entity will enter the next step of the knowledge graph processing; otherwise, entities with lower evaluation values ​​will be discarded to avoid irrelevant or low-quality entities from polluting the knowledge graph.

[0054] Based on the solution of the present invention, the accuracy and reliability of the knowledge graph construction in the field of engineering machinery are significantly improved by performing weighted sum operations on the probability of occurrence and the average similarity. The weighting mechanism makes the entity screening process more flexible, and the weights can be adjusted according to different application scenarios to meet diverse needs. The entity evaluation value after weighted summation provides a comprehensive quality assessment standard to ensure that the entities that ultimately enter the knowledge graph are highly semantically relevant and have practical application value. This screening mechanism not only improves the accuracy of the knowledge graph, but also improves its effectiveness in intelligent retrieval, fault diagnosis, and decision support, ultimately providing strong support for the intelligent development of the field of engineering machinery.

[0055] Step S40: performing an update of the engineering machinery knowledge graph based on the retained identified entities and their corresponding similar entities.

[0056] Specifically, when it is determined that there is no corresponding identification entity in the pre-constructed engineering machinery knowledge graph, the correlation between each identification entity and the corresponding similar entity and the corresponding entity type in the pre-constructed engineering machinery knowledge graph is calculated; classification is performed on each identification entity and the corresponding similar entity based on the correlation, and graph nodes are constructed for each identification entity and the corresponding similar entity based on the classification results; after it is determined that the graph node meets the node standard, the engineering machinery knowledge graph is updated based on the constructed graph node.

[0057] Furthermore, classification is performed on each identified entity and the corresponding similar entity based on the correlation, including: determining the entity type of the corresponding identified entity / similar entity based on the entity type corresponding to the maximum value of the correlation between each identified entity / similar entity and each entity type in the pre-built engineering machinery knowledge graph; wherein the entity type is a node entity or a node attribute.

[0058] Furthermore, the graph node construction is performed on each identified entity and the corresponding similar entity based on the classification result, including: calculating the correlation between each node entity and the corresponding node attribute, and taking the combination with a correlation greater than a preset correlation threshold as a node combination; extracting a node construction template, performing graph node construction based on each node combination, and obtaining multiple candidate graph nodes.

[0059] Furthermore, after determining whether the graph node meets the node standard, the construction machinery knowledge graph is updated based on the constructed graph node, including: determining whether each candidate graph node has missing information based on the node construction template; if there is missing information, generating the missing information based on the existing information and the pre-trained node generation model, and updating each candidate graph node based on the generated missing information; treating the candidate graph nodes that meet the node construction template as standard graph nodes, and updating the construction machinery knowledge graph based on the standard graph nodes.

[0060] In an embodiment of the present invention, when new identified entities and similar entities are identified, the system needs to check whether these entities already exist in the pre-built engineering machinery knowledge graph. If an identified entity or its similar entities have not yet been included in the knowledge graph, the system will start an update process. In this process, the system first calculates the correlation between these newly identified entities and their similar entities and the existing entity types in the knowledge graph. The correlation calculation is based on semantic or statistical methods, aiming to determine the strength of association between these new entities and existing graph entities. Once the correlation is calculated, the system will classify the newly identified entities and their similar entities according to the correlation. The goal of classification is to determine the role of these entities in the knowledge graph-are they independent nodes in the graph, or are they attributes of other nodes. Specifically, the system determines the entity type based on the maximum value of the correlation between the new entity and the existing entity type. For example, if an entity has the highest correlation with a node entity of a certain type, it will be classified as a node of that type; if it has a higher correlation with the attribute of a certain node, it will be classified as a node attribute.

[0061] Furthermore, after the classification is completed, the system will enter the node construction stage. Node construction includes calculating the correlation between each node entity and the corresponding node attributes, and taking the combination with a correlation greater than a preset threshold as a node combination. Then, the system will extract the node construction template, perform graph node construction based on these combinations, and finally obtain multiple candidate graph nodes. This stage ensures that the newly added nodes have a reasonable association with other parts of the graph, thereby maintaining the structural integrity of the graph.

[0062] Furthermore, after the candidate graph nodes are constructed, the system will verify these nodes to determine whether they meet the node construction standards. The verification includes checking whether the nodes have missing information, especially those attributes that are crucial in knowledge graph applications. For example, if a node lacks key attributes such as fault symptoms, the system will use a pre-trained node generation model to supplement the missing information. By generating a model, the system can generate reasonable supplementary data based on existing information, making the node more complete and reliable. When all nodes are verified and supplemented, the system will mark the candidate graph nodes that meet the standards as standard graph nodes. Finally, these standard nodes will be integrated into the knowledge graph to complete the update of the graph.

[0063] In one possible embodiment, if the system identifies the entity "arm" and finds through calculation that its correlation with the existing node "one arm" is 0.9, then the system will classify "arm" as an attribute of "one arm". If the entity category of "arm" is determined to be "part", the system will combine it with "one arm" to form a node combination, such as {"one arm": {"part": "arm"}}. In the next step, the system will check whether there is missing information in the node combination. For example, if the "fault phenomenon" attribute of the node is empty, the system will use the generative model to supplement this information to make it {"one arm": {"part": "arm", "fault phenomenon": "drop arm"}}. In this way, the node can be added to the knowledge graph.

[0064] Based on the scheme of the present invention, through detailed entity classification and node construction steps, the system can effectively identify and fill gaps in the knowledge graph. In particular, by using a generative model to supplement missing information, the integrity and reliability of the graph are ensured. Enhance the semantic relevance of the graph: By calculating the relevance of entities to existing entity types in the graph, the system can ensure that newly added nodes have reasonable semantic relevance to the existing structure. This relevance improves the semantic level of the knowledge graph, enabling it to more accurately support complex query and reasoning tasks.

[0065] Furthermore, the updated knowledge graph is not only more structurally sound, but also better able to support intelligent applications in the field of engineering machinery, such as equipment failure prediction, maintenance plan optimization, and operation recommendations. These applications rely on high-quality knowledge graphs, and this method significantly improves the practicality and application value of the graph through a strict update process. Another important technical effect of this method is its flexibility. Through adjustable relevance thresholds and weight settings, the system can adjust the graph update strategy according to the needs of different application scenarios, thereby maximizing the effectiveness of the graph.

[0066] Figure 3: is a system structure diagram of an engineering machinery knowledge graph data enhancement system provided by an embodiment of the present invention. Figure 3 As shown, an embodiment of the present invention provides a system for enhancing data of a knowledge graph of engineering machinery, the system comprising: a collection unit, for collecting text information of engineering machinery to be stored in the database, identifying engineering machinery-related entities based on the text information and generating similar entities based on each identified entity, so as to obtain multiple similar entities of each identified entity, and construct a similar entity set corresponding to each identified entity; a similarity determination unit, for filtering out similar entities whose similarity is less than a preset similarity threshold based on the similarity between each similar entity and the corresponding identified entity in the similar entity set, and calculating the average similarity of the corresponding identified entities based on the remaining similar entities; an evaluation unit, for calculating the probability of occurrence of each identified entity in the text information based on the word embedding technology, and performing a weighted sum operation based on the occurrence probability and the average similarity of the corresponding identified entity to obtain an entity evaluation value, and retaining the identified entities whose entity evaluation values ​​are greater than the preset evaluation value threshold and their corresponding similar entities; an updating unit, for updating the knowledge graph of engineering machinery based on the retained identified entities and their corresponding similar entities.

[0067] An embodiment of the present invention also provides a computer-readable storage medium, which stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned engineering machinery knowledge graph data enhancement method.

[0068] Those skilled in the art will understand that all or part of the steps in the method for implementing the above-mentioned embodiments can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including several instructions for making a single-chip microcomputer, a chip or a processor (processor) perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0069] The optional embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the technical concept of the embodiments of the present invention, the technical scheme of the embodiments of the present invention can be subjected to a variety of simple modifications, and these simple modifications all belong to the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the embodiments of the present invention will not further describe various possible combinations.

[0070] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the embodiments of the present invention, they should also be regarded as the contents disclosed in the embodiments of the present invention.

Claims

1. A method for data enhancement of engineering machinery knowledge graph, characterized in that: The method comprises: Collect text information of engineering machinery to be stored, identify engineering machinery related entities based on the text information, generate similar entities based on each identified entity, and construct a similar entity set corresponding to each identified entity; Calculate the average similarity of the corresponding identified entities based on the similar entity sets corresponding to each identified entity; Calculate the occurrence probability of each recognized entity in the text information based on the word embedding technology, and obtain the entity evaluation value based on the occurrence probability and the average similarity of the corresponding recognized entity, and retain the recognized entity whose entity evaluation value is greater than a preset evaluation value threshold and its corresponding similar entity; The engineering machinery knowledge graph is updated based on the retained identified entities and their corresponding similar entities.

2. The method according to claim 1, characterized in that The identifying of engineering machinery-related entities based on the text information includes: Perform text cleaning on the text information, and perform word segmentation processing after text cleaning to obtain multiple word segmentation items; wherein, Based on the pre-built engineering machinery knowledge graph, the corresponding engineering machinery entity recognition model is loaded, and based on the loaded engineering machinery entity recognition model, engineering machinery related entities are recognized in each participle item to obtain recognized entities.

3. The method according to claim 1, characterized in that The rules for generating similar entities based on each identified entity are: S1) taking each recognized entity as an input item, loading a pre-trained engineering machinery generation model, obtaining multiple generation entities, and assigning a corresponding similarity probability to each generation entity; S2) Compare the similarity probabilities of each generated entity, and take the previously generated entity with the largest similarity probability as the similar entity for this round of training: S3) Deleting the determined similar entities from the entities to be generated, and repeating steps S1)-S2) N times to obtain N similar entities.

4. The method according to claim 1, characterized in that: The calculating the average similarity of the corresponding identified entities based on the similar entity sets corresponding to the identified entities includes: Based on the similarity between each similar entity in the similar entity set and the corresponding identified entity, similar entities whose similarity is less than a preset similarity threshold are filtered out, and the average similarity of the corresponding identified entity is calculated based on the remaining similar entities; wherein, Based on the similarity between each similar entity in the similar entity set and the corresponding identified entity, similar entities whose similarity is less than a preset similarity threshold are filtered out, including: Convert each similar entity and the identified entity in the similar entity set into a vector of the same preset dimension; Calculate the similarity between the original entity and each generated similar entity based on cosine similarity; The magnitude relationship between the similarity corresponding to each similar entity and a preset similarity threshold is compared, and similar entities whose similarity threshold is less than the preset similarity threshold are filtered.

5. The method according to claim 1, characterized in that The occurrence probability of each recognized entity in the text information is calculated based on the word embedding technology, including: Performing word segmentation processing on the text information to determine the location of each recognized entity in the text information to obtain context word entities; Load the pre-trained word embedding technology to process each context word entity and the corresponding recognition entity into a vector of preset dimensions; Based on each context word entity vector and each corresponding recognition entity vector, the association degree between each recognition entity and each context word is calculated respectively; Based on the pre-built engineering machinery knowledge graph, count the occurrence frequency of each recognized entity in the text type of the text information; Calculate the occurrence probability of each recognized entity in the text information based on the association degree and the occurrence frequency; Based on the word embedding technology, the method further comprises: calculating the occurrence probability of each recognized entity in the text information; Filter out the identified entities whose occurrence probability is less than the preset probability threshold, as well as the similar entities of the corresponding identified entities.

6. The method according to claim 1, characterized in that Obtaining an entity evaluation value based on the occurrence probability and the average similarity of the corresponding identified entity, including: A weighted sum operation is performed based on the occurrence probability and the average similarity of the corresponding identified entity to obtain an entity evaluation value, wherein: A weighted sum operation is performed based on the occurrence probability and the average similarity of the corresponding identified entities, including: Weighting the occurrence probability and average similarity of each identified entity based on a preset weighting scheme; All identified entities are traversed, and weighted sum operations are performed based on the weighted occurrence probability and the weighted average similarity to obtain the entity evaluation value corresponding to each identified entity.

7. The method according to claim 1, characterized in that The updating of the engineering machinery knowledge graph based on the retained identification entities and their corresponding similar entities includes: When it is determined that there is no corresponding identification entity in the pre-constructed engineering machinery knowledge graph, the correlation between each identification entity and the corresponding similar entity and the corresponding entity type in the pre-constructed engineering machinery knowledge graph is calculated; Classify each identified entity and the corresponding similar entity based on the correlation, and construct graph nodes for each identified entity and the corresponding similar entity based on the classification results; After determining that the graph node meets the node standard, the engineering machinery knowledge graph is updated based on the constructed graph node.

8. The method according to claim 7, characterized in that Classifying each identified entity and the corresponding similar entity based on the correlation includes: Based on the entity type corresponding to the maximum value of the correlation between each identified entity / similar entity and each entity type in the pre-constructed engineering machinery knowledge graph, the entity type corresponding to the identified entity / similar entity is determined; wherein, The entity type is either a node entity or a node attribute.

9. The method according to claim 8, characterized in that The step of constructing graph nodes for each identified entity and corresponding similar entities based on the classification results includes: Calculate the correlation between each node entity and the corresponding node attribute, and take the combination with a correlation greater than a preset correlation threshold as a node combination; Extract node construction templates, execute graph node construction based on the combination of nodes, and obtain multiple candidate graph nodes.

10. The method according to claim 9, characterized in that After determining that the graph node meets the node standard, executing the engineering machinery knowledge graph update based on the constructed graph node includes: Determine whether each candidate graph node has missing information based on the node construction template; If there is missing information, the missing information is generated based on the existing information and the pre-trained node generation model, and each candidate graph node is updated based on the generated missing information; The candidate graph nodes that meet the node construction template are used as standard graph nodes, and the engineering machinery knowledge graph is updated based on the standard graph nodes.

11. A construction machinery knowledge graph data enhancement system, characterized in that: The system comprises: A collection unit, used to collect text information of engineering machinery to be stored, identify engineering machinery related entities based on the text information, generate similar entities based on each identified entity, and construct a similar entity set corresponding to each identified entity; A similarity determination unit, configured to calculate an average similarity of corresponding identified entities based on a similar entity set corresponding to each identified entity; An evaluation unit, used to calculate the occurrence probability of each recognized entity in the text information based on the word embedding technology, and obtain an entity evaluation value based on the occurrence probability and the average similarity of the corresponding recognized entity, and retain the recognized entity whose entity evaluation value is greater than a preset evaluation value threshold and its corresponding similar entity; An updating unit is used to perform an update of the engineering machinery knowledge graph based on the retained identified entities and their corresponding similar entities.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the engineering machinery knowledge graph data enhancement method described in any one of claims 1 to 10.

Images